Image segmentation by non-topological erosion and topological expansion

A new segmentation method is suggested to distinguish the foreground from the background in gray-level images. The method is based on a 2-step process, respectively employing non-topological pixel removal (non-topological erosion) and topological region growing (topological expansion). The first step is aimed at identifying suitable seeds, corresponding to the objects of interest in the image, while the second step associates to the identified seeds pixels removed during the first step, provided that fusions are not created. Segmentation is accomplished by using also information derived from a lower resolution representation of the image, with the purpose of reducing the number of foreground components to the most significant ones. Some hints regarding extension of the method to color images are also discussed.